Five Questions to Ask Any AI Service Provider Before You Sign Up

The market for AI services aimed at small businesses has grown faster than the quality controls on it. For every provider that operates a genuine managed service with real oversight and accountability, there are several that deliver a self-configure software tool with a premium-sounding description. At the pitch stage, the two often sound identical.

The difference only becomes apparent later — usually when something goes wrong, when the business changes and the workflow does not, or when the business owner realises they are maintaining the system themselves despite paying for a service.

These five questions are designed to surface that difference before you sign up.

In short: A genuine managed service takes responsibility for the output and maintains the workflow over time. A self-serve tool with a managed-service label does not. These questions will tell you which one you are looking at.

1. Who is responsible for the accuracy of the output?

This is the most direct question, and the answer reveals the most.

A genuine managed service will give you a clear answer: "We are. If the workflow produces an inaccurate response, we fix it. You tell us when something is wrong and we update it." The responsibility for the knowledge base, the quality of responses, and the ongoing calibration sits with the provider.

A self-serve tool positioned as a managed service will shift the responsibility back to you: "It depends on how you've configured it." Or: "The AI generates responses based on the information you provided." Or: "You'll need to review and update the system as your business changes."

These are not the same thing. In the first case, you are buying a service that takes accountability for its output. In the second, you are renting software that runs on your behalf but is ultimately your problem to maintain.

If the provider cannot answer "us" to this question, the rest of the sales process is premature.

2. What happens when the workflow produces a wrong answer?

Every workflow will eventually produce an answer that is not quite right: a pricing figure that has since changed, a service detail that was not fully captured in the knowledge base, an unusual situation that was not anticipated during setup. This is not a failure — it is inevitable. The question is what the process is when it happens.

Ask the provider: "If the workflow gives an incorrect or off-brand response to a prospect, what happens?" The answer should include:

• A human review process that catches issues before they escalate
• A clear update mechanism that fixes the underlying knowledge base
• A timeline for how quickly corrections are applied

If the answer involves waiting for you to notice the problem yourself and then reconfiguring the tool, the monitoring and accountability is yours, not theirs. A managed service monitors its own output. A software tool runs until you check it.

3. How does the workflow handle sensitive or unusual enquiries?

Not every enquiry should be handled automatically. A complaint. An unusually high-value opportunity that needs a personalised response. An enquirer who is clearly distressed. A situation the workflow has never encountered. How these are handled is a direct test of whether a service has genuine human oversight built in.

A well-designed managed service will have specific escalation triggers: categories of enquiry that go to a human reviewer before any response is sent. The provider should be able to describe what those triggers are and what the handover process looks like.

If the answer is "the AI will handle it" or "it will redirect to a generic fallback message", ask what happens in a case where the fallback is insufficient. A provider without a real escalation process is running an automated system with no catch for the cases that most need human judgement.

4. How does the workflow get updated when our business changes?

Businesses change constantly. Services are added. Areas are expanded. Pricing changes. A key team member moves on. The workflow does not know about any of this unless someone updates it.

The question to ask is: "When we change our pricing or add a new service, how does that get into the workflow?" The answer should be simple and quick — you notify the provider, they update the knowledge base, and it is live within a defined timeframe (typically 24-48 hours).

If the answer is "you log into the dashboard and update the settings", or "you edit the prompt in the configuration panel", the maintenance is yours. That is not a managed service. It is a configurable tool, and configuring it takes time, attention, and expertise that most small business owners are buying the service specifically to avoid spending.

The update process is where self-serve tools break down most often. The business changes, the update does not get made, and the workflow keeps answering with stale information for weeks or months until someone notices.

5. What does the contract or commitment look like?

A provider who is confident in their service will not require a long-term contract to make the economics work. The pilot model — a short initial period with a money-back guarantee on setup, then month-to-month thereafter — is the sign of a provider who believes in the results enough to let the results sell the continuation.

Long minimum terms (six months, twelve months, annual prepayment) on a product the business has never used before should prompt the question: why? The usual answer is that the provider needs the contract length because the return on the relationship takes a long time to materialise — either because setup is complex and front-loaded, or because they expect attrition and are smoothing the revenue. Neither of those should be the business owner's problem.

Ask: "What is the minimum commitment? What happens if the workflow is not live within the agreed setup window? What is the cancellation process?"

A genuine managed service answers all three questions easily. A provider who struggles with any of them should be asked to explain further before any agreement is signed.

The pattern these questions reveal

A genuine managed service is accountable for its output, monitors what it produces, escalates what requires human judgement, updates when the business changes, and operates on terms that reflect confidence in the results. The questions above test each of those properties directly.

Most providers in this space will answer some of them well and some of them poorly. The combination of answers tells you more than any pitch deck, case study, or feature comparison.

Frequently asked questions

Are there other questions I should ask besides these five?

These five cover the core accountability and service quality questions. Depending on the business, additional questions about data privacy (where is my customer data stored? what is your data retention policy?), integrations (which email systems and CRMs do you connect with?), and industry experience (have you worked with businesses in our sector?) may also be relevant.

What if the provider wants to show me a demo before I ask these questions?

Watch the demo. Then ask the questions. A demo shows what the product can do in ideal conditions — these questions test what happens when conditions are not ideal.

Should I get the answers to these questions in writing?

For the key points — particularly around who is responsible for output accuracy and what the setup guarantee covers — yes. Not because providers are dishonest, but because verbal commitments made at the pitch stage by someone who does not run the service can diverge from what the operations team actually delivers. A clear written summary protects both sides.

How do I evaluate a provider I have not been able to speak with in detail?

Look at what the product requires from you. If the setup process requires you to configure the system, write the prompts, and test the outputs, it is a self-serve tool regardless of how it is described. If the provider does the setup, tests it themselves, and reviews the output on an ongoing basis, it is a managed service. The amount of work placed on you at setup and during operation is the most reliable signal. For a concrete picture of what a managed service looks like in practice, what happens in the first 30 days of a managed enquiry workflow gives a step-by-step breakdown from sign-off to go-live.